1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
|
/*
* Copyright (C) 2016 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package android.util;
import android.annotation.HalfFloat;
import android.annotation.NonNull;
import android.annotation.Nullable;
/**
* <p>The {@code Half} class is a wrapper and a utility class to manipulate half-precision 16-bit
* <a href="https://en.wikipedia.org/wiki/Half-precision_floating-point_format">IEEE 754</a>
* floating point data types (also called fp16 or binary16). A half-precision float can be
* created from or converted to single-precision floats, and is stored in a short data type.
* To distinguish short values holding half-precision floats from regular short values,
* it is recommended to use the <code>@HalfFloat</code> annotation.</p>
*
* <p>The IEEE 754 standard specifies an fp16 as having the following format:</p>
* <ul>
* <li>Sign bit: 1 bit</li>
* <li>Exponent width: 5 bits</li>
* <li>Significand: 10 bits</li>
* </ul>
*
* <p>The format is laid out as follows:</p>
* <pre>
* 1 11111 1111111111
* ^ --^-- -----^----
* sign | |_______ significand
* |
* -- exponent
* </pre>
*
* <p>Half-precision floating points can be useful to save memory and/or
* bandwidth at the expense of range and precision when compared to single-precision
* floating points (fp32).</p>
* <p>To help you decide whether fp16 is the right storage type for you need, please
* refer to the table below that shows the available precision throughout the range of
* possible values. The <em>precision</em> column indicates the step size between two
* consecutive numbers in a specific part of the range.</p>
*
* <table summary="Precision of fp16 across the range">
* <tr><th>Range start</th><th>Precision</th></tr>
* <tr><td>0</td><td>1 ⁄ 16,777,216</td></tr>
* <tr><td>1 ⁄ 16,384</td><td>1 ⁄ 16,777,216</td></tr>
* <tr><td>1 ⁄ 8,192</td><td>1 ⁄ 8,388,608</td></tr>
* <tr><td>1 ⁄ 4,096</td><td>1 ⁄ 4,194,304</td></tr>
* <tr><td>1 ⁄ 2,048</td><td>1 ⁄ 2,097,152</td></tr>
* <tr><td>1 ⁄ 1,024</td><td>1 ⁄ 1,048,576</td></tr>
* <tr><td>1 ⁄ 512</td><td>1 ⁄ 524,288</td></tr>
* <tr><td>1 ⁄ 256</td><td>1 ⁄ 262,144</td></tr>
* <tr><td>1 ⁄ 128</td><td>1 ⁄ 131,072</td></tr>
* <tr><td>1 ⁄ 64</td><td>1 ⁄ 65,536</td></tr>
* <tr><td>1 ⁄ 32</td><td>1 ⁄ 32,768</td></tr>
* <tr><td>1 ⁄ 16</td><td>1 ⁄ 16,384</td></tr>
* <tr><td>1 ⁄ 8</td><td>1 ⁄ 8,192</td></tr>
* <tr><td>1 ⁄ 4</td><td>1 ⁄ 4,096</td></tr>
* <tr><td>1 ⁄ 2</td><td>1 ⁄ 2,048</td></tr>
* <tr><td>1</td><td>1 ⁄ 1,024</td></tr>
* <tr><td>2</td><td>1 ⁄ 512</td></tr>
* <tr><td>4</td><td>1 ⁄ 256</td></tr>
* <tr><td>8</td><td>1 ⁄ 128</td></tr>
* <tr><td>16</td><td>1 ⁄ 64</td></tr>
* <tr><td>32</td><td>1 ⁄ 32</td></tr>
* <tr><td>64</td><td>1 ⁄ 16</td></tr>
* <tr><td>128</td><td>1 ⁄ 8</td></tr>
* <tr><td>256</td><td>1 ⁄ 4</td></tr>
* <tr><td>512</td><td>1 ⁄ 2</td></tr>
* <tr><td>1,024</td><td>1</td></tr>
* <tr><td>2,048</td><td>2</td></tr>
* <tr><td>4,096</td><td>4</td></tr>
* <tr><td>8,192</td><td>8</td></tr>
* <tr><td>16,384</td><td>16</td></tr>
* <tr><td>32,768</td><td>32</td></tr>
* </table>
*
* <p>This table shows that numbers higher than 1024 lose all fractional precision.</p>
*/
@SuppressWarnings("SimplifiableIfStatement")
public final class Half extends Number implements Comparable<Half> {
/**
* The number of bits used to represent a half-precision float value.
*/
public static final int SIZE = 16;
/**
* Epsilon is the difference between 1.0 and the next value representable
* by a half-precision floating-point.
*/
public static final @HalfFloat short EPSILON = (short) 0x1400;
/**
* Maximum exponent a finite half-precision float may have.
*/
public static final int MAX_EXPONENT = 15;
/**
* Minimum exponent a normalized half-precision float may have.
*/
public static final int MIN_EXPONENT = -14;
/**
* Smallest negative value a half-precision float may have.
*/
public static final @HalfFloat short LOWEST_VALUE = (short) 0xfbff;
/**
* Maximum positive finite value a half-precision float may have.
*/
public static final @HalfFloat short MAX_VALUE = (short) 0x7bff;
/**
* Smallest positive normal value a half-precision float may have.
*/
public static final @HalfFloat short MIN_NORMAL = (short) 0x0400;
/**
* Smallest positive non-zero value a half-precision float may have.
*/
public static final @HalfFloat short MIN_VALUE = (short) 0x0001;
/**
* A Not-a-Number representation of a half-precision float.
*/
public static final @HalfFloat short NaN = (short) 0x7e00;
/**
* Negative infinity of type half-precision float.
*/
public static final @HalfFloat short NEGATIVE_INFINITY = (short) 0xfc00;
/**
* Negative 0 of type half-precision float.
*/
public static final @HalfFloat short NEGATIVE_ZERO = (short) 0x8000;
/**
* Positive infinity of type half-precision float.
*/
public static final @HalfFloat short POSITIVE_INFINITY = (short) 0x7c00;
/**
* Positive 0 of type half-precision float.
*/
public static final @HalfFloat short POSITIVE_ZERO = (short) 0x0000;
private static final int FP16_SIGN_SHIFT = 15;
private static final int FP16_SIGN_MASK = 0x8000;
private static final int FP16_EXPONENT_SHIFT = 10;
private static final int FP16_EXPONENT_MASK = 0x1f;
private static final int FP16_SIGNIFICAND_MASK = 0x3ff;
private static final int FP16_EXPONENT_BIAS = 15;
private static final int FP16_COMBINED = 0x7fff;
private static final int FP16_EXPONENT_MAX = 0x7c00;
private static final int FP32_SIGN_SHIFT = 31;
private static final int FP32_EXPONENT_SHIFT = 23;
private static final int FP32_EXPONENT_MASK = 0xff;
private static final int FP32_SIGNIFICAND_MASK = 0x7fffff;
private static final int FP32_EXPONENT_BIAS = 127;
private static final int FP32_QNAN_MASK = 0x400000;
private static final int FP32_DENORMAL_MAGIC = 126 << 23;
private static final float FP32_DENORMAL_FLOAT = Float.intBitsToFloat(FP32_DENORMAL_MAGIC);
private final @HalfFloat short mValue;
/**
* Constructs a newly allocated {@code Half} object that represents the
* half-precision float type argument.
*
* @param value The value to be represented by the {@code Half}
*/
public Half(@HalfFloat short value) {
mValue = value;
}
/**
* Constructs a newly allocated {@code Half} object that represents the
* argument converted to a half-precision float.
*
* @param value The value to be represented by the {@code Half}
*
* @see #toHalf(float)
*/
public Half(float value) {
mValue = toHalf(value);
}
/**
* Constructs a newly allocated {@code Half} object that
* represents the argument converted to a half-precision float.
*
* @param value The value to be represented by the {@code Half}
*
* @see #toHalf(float)
*/
public Half(double value) {
mValue = toHalf((float) value);
}
/**
* <p>Constructs a newly allocated {@code Half} object that represents the
* half-precision float value represented by the string.
* The string is converted to a half-precision float value as if by the
* {@link #valueOf(String)} method.</p>
*
* <p>Calling this constructor is equivalent to calling:</p>
* <pre>
* new Half(Float.parseFloat(value))
* </pre>
*
* @param value A string to be converted to a {@code Half}
* @throws NumberFormatException if the string does not contain a parsable number
*
* @see Float#valueOf(java.lang.String)
* @see #toHalf(float)
*/
public Half(@NonNull String value) throws NumberFormatException {
mValue = toHalf(Float.parseFloat(value));
}
/**
* Returns the half-precision value of this {@code Half} as a {@code short}
* containing the bit representation described in {@link Half}.
*
* @return The half-precision float value represented by this object
*/
public @HalfFloat short halfValue() {
return mValue;
}
/**
* Returns the value of this {@code Half} as a {@code byte} after
* a narrowing primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code byte}
*/
@Override
public byte byteValue() {
return (byte) toFloat(mValue);
}
/**
* Returns the value of this {@code Half} as a {@code short} after
* a narrowing primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code short}
*/
@Override
public short shortValue() {
return (short) toFloat(mValue);
}
/**
* Returns the value of this {@code Half} as a {@code int} after
* a narrowing primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code int}
*/
@Override
public int intValue() {
return (int) toFloat(mValue);
}
/**
* Returns the value of this {@code Half} as a {@code long} after
* a narrowing primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code long}
*/
@Override
public long longValue() {
return (long) toFloat(mValue);
}
/**
* Returns the value of this {@code Half} as a {@code float} after
* a widening primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code float}
*/
@Override
public float floatValue() {
return toFloat(mValue);
}
/**
* Returns the value of this {@code Half} as a {@code double} after
* a widening primitive conversion.
*
* @return The half-precision float value represented by this object
* converted to type {@code double}
*/
@Override
public double doubleValue() {
return toFloat(mValue);
}
/**
* Returns true if this {@code Half} value represents a Not-a-Number,
* false otherwise.
*
* @return True if the value is a NaN, false otherwise
*/
public boolean isNaN() {
return isNaN(mValue);
}
/**
* Compares this object against the specified object. The result is {@code true}
* if and only if the argument is not {@code null} and is a {@code Half} object
* that represents the same half-precision value as the this object. Two
* half-precision values are considered to be the same if and only if the method
* {@link #halfToIntBits(short)} returns an identical {@code int} value for both.
*
* @param o The object to compare
* @return True if the objects are the same, false otherwise
*
* @see #halfToIntBits(short)
*/
@Override
public boolean equals(@Nullable Object o) {
return (o instanceof Half) &&
(halfToIntBits(((Half) o).mValue) == halfToIntBits(mValue));
}
/**
* Returns a hash code for this {@code Half} object. The result is the
* integer bit representation, exactly as produced by the method
* {@link #halfToIntBits(short)}, of the primitive half-precision float
* value represented by this {@code Half} object.
*
* @return A hash code value for this object
*/
@Override
public int hashCode() {
return hashCode(mValue);
}
/**
* Returns a string representation of the specified half-precision
* float value. See {@link #toString(short)} for more information.
*
* @return A string representation of this {@code Half} object
*/
@NonNull
@Override
public String toString() {
return toString(mValue);
}
/**
* <p>Compares the two specified half-precision float values. The following
* conditions apply during the comparison:</p>
*
* <ul>
* <li>{@link #NaN} is considered by this method to be equal to itself and greater
* than all other half-precision float values (including {@code #POSITIVE_INFINITY})</li>
* <li>{@link #POSITIVE_ZERO} is considered by this method to be greater than
* {@link #NEGATIVE_ZERO}.</li>
* </ul>
*
* @param h The half-precision float value to compare to the half-precision value
* represented by this {@code Half} object
*
* @return The value {@code 0} if {@code x} is numerically equal to {@code y}; a
* value less than {@code 0} if {@code x} is numerically less than {@code y};
* and a value greater than {@code 0} if {@code x} is numerically greater
* than {@code y}
*/
@Override
public int compareTo(@NonNull Half h) {
return compare(mValue, h.mValue);
}
/**
* Returns a hash code for a half-precision float value.
*
* @param h The value to hash
*
* @return A hash code value for a half-precision float value
*/
public static int hashCode(@HalfFloat short h) {
return halfToIntBits(h);
}
/**
* <p>Compares the two specified half-precision float values. The following
* conditions apply during the comparison:</p>
*
* <ul>
* <li>{@link #NaN} is considered by this method to be equal to itself and greater
* than all other half-precision float values (including {@code #POSITIVE_INFINITY})</li>
* <li>{@link #POSITIVE_ZERO} is considered by this method to be greater than
* {@link #NEGATIVE_ZERO}.</li>
* </ul>
*
* @param x The first half-precision float value to compare.
* @param y The second half-precision float value to compare
*
* @return The value {@code 0} if {@code x} is numerically equal to {@code y}, a
* value less than {@code 0} if {@code x} is numerically less than {@code y},
* and a value greater than {@code 0} if {@code x} is numerically greater
* than {@code y}
*/
public static int compare(@HalfFloat short x, @HalfFloat short y) {
if (less(x, y)) return -1;
if (greater(x, y)) return 1;
// Collapse NaNs, akin to halfToIntBits(), but we want to keep
// (signed) short value types to preserve the ordering of -0.0
// and +0.0
short xBits = (x & FP16_COMBINED) > FP16_EXPONENT_MAX ? NaN : x;
short yBits = (y & FP16_COMBINED) > FP16_EXPONENT_MAX ? NaN : y;
return (xBits == yBits ? 0 : (xBits < yBits ? -1 : 1));
}
/**
* <p>Returns a representation of the specified half-precision float value
* according to the bit layout described in {@link Half}.</p>
*
* <p>Similar to {@link #halfToIntBits(short)}, this method collapses all
* possible Not-a-Number values to a single canonical Not-a-Number value
* defined by {@link #NaN}.</p>
*
* @param h A half-precision float value
* @return The bits that represent the half-precision float value
*
* @see #halfToIntBits(short)
*/
public static @HalfFloat short halfToShortBits(@HalfFloat short h) {
return (h & FP16_COMBINED) > FP16_EXPONENT_MAX ? NaN : h;
}
/**
* <p>Returns a representation of the specified half-precision float value
* according to the bit layout described in {@link Half}.</p>
*
* <p>Unlike {@link #halfToRawIntBits(short)}, this method collapses all
* possible Not-a-Number values to a single canonical Not-a-Number value
* defined by {@link #NaN}.</p>
*
* @param h A half-precision float value
* @return The bits that represent the half-precision float value
*
* @see #halfToRawIntBits(short)
* @see #halfToShortBits(short)
* @see #intBitsToHalf(int)
*/
public static int halfToIntBits(@HalfFloat short h) {
return (h & FP16_COMBINED) > FP16_EXPONENT_MAX ? NaN : h & 0xffff;
}
/**
* <p>Returns a representation of the specified half-precision float value
* according to the bit layout described in {@link Half}.</p>
*
* <p>The argument is considered to be a representation of a half-precision
* float value according to the bit layout described in {@link Half}. The 16
* most significant bits of the returned value are set to 0.</p>
*
* @param h A half-precision float value
* @return The bits that represent the half-precision float value
*
* @see #halfToIntBits(short)
* @see #intBitsToHalf(int)
*/
public static int halfToRawIntBits(@HalfFloat short h) {
return h & 0xffff;
}
/**
* <p>Returns the half-precision float value corresponding to a given
* bit representation.</p>
*
* <p>The argument is considered to be a representation of a half-precision
* float value according to the bit layout described in {@link Half}. The 16
* most significant bits of the argument are ignored.</p>
*
* @param bits An integer
* @return The half-precision float value with the same bit pattern
*/
public static @HalfFloat short intBitsToHalf(int bits) {
return (short) (bits & 0xffff);
}
/**
* Returns the first parameter with the sign of the second parameter.
* This method treats NaNs as having a sign.
*
* @param magnitude A half-precision float value providing the magnitude of the result
* @param sign A half-precision float value providing the sign of the result
* @return A value with the magnitude of the first parameter and the sign
* of the second parameter
*/
public static @HalfFloat short copySign(@HalfFloat short magnitude, @HalfFloat short sign) {
return (short) ((sign & FP16_SIGN_MASK) | (magnitude & FP16_COMBINED));
}
/**
* Returns the absolute value of the specified half-precision float.
* Special values are handled in the following ways:
* <ul>
* <li>If the specified half-precision float is NaN, the result is NaN</li>
* <li>If the specified half-precision float is zero (negative or positive),
* the result is positive zero (see {@link #POSITIVE_ZERO})</li>
* <li>If the specified half-precision float is infinity (negative or positive),
* the result is positive infinity (see {@link #POSITIVE_INFINITY})</li>
* </ul>
*
* @param h A half-precision float value
* @return The absolute value of the specified half-precision float
*/
public static @HalfFloat short abs(@HalfFloat short h) {
return (short) (h & FP16_COMBINED);
}
/**
* Returns the closest integral half-precision float value to the specified
* half-precision float value. Special values are handled in the
* following ways:
* <ul>
* <li>If the specified half-precision float is NaN, the result is NaN</li>
* <li>If the specified half-precision float is infinity (negative or positive),
* the result is infinity (with the same sign)</li>
* <li>If the specified half-precision float is zero (negative or positive),
* the result is zero (with the same sign)</li>
* </ul>
*
* @param h A half-precision float value
* @return The value of the specified half-precision float rounded to the nearest
* half-precision float value
*/
public static @HalfFloat short round(@HalfFloat short h) {
int bits = h & 0xffff;
int e = bits & 0x7fff;
int result = bits;
if (e < 0x3c00) {
result &= FP16_SIGN_MASK;
result |= (0x3c00 & (e >= 0x3800 ? 0xffff : 0x0));
} else if (e < 0x6400) {
e = 25 - (e >> 10);
int mask = (1 << e) - 1;
result += (1 << (e - 1));
result &= ~mask;
}
return (short) result;
}
/**
* Returns the smallest half-precision float value toward negative infinity
* greater than or equal to the specified half-precision float value.
* Special values are handled in the following ways:
* <ul>
* <li>If the specified half-precision float is NaN, the result is NaN</li>
* <li>If the specified half-precision float is infinity (negative or positive),
* the result is infinity (with the same sign)</li>
* <li>If the specified half-precision float is zero (negative or positive),
* the result is zero (with the same sign)</li>
* </ul>
*
* @param h A half-precision float value
* @return The smallest half-precision float value toward negative infinity
* greater than or equal to the specified half-precision float value
*/
public static @HalfFloat short ceil(@HalfFloat short h) {
int bits = h & 0xffff;
int e = bits & 0x7fff;
int result = bits;
if (e < 0x3c00) {
result &= FP16_SIGN_MASK;
result |= 0x3c00 & -(~(bits >> 15) & (e != 0 ? 1 : 0));
} else if (e < 0x6400) {
e = 25 - (e >> 10);
int mask = (1 << e) - 1;
result += mask & ((bits >> 15) - 1);
result &= ~mask;
}
return (short) result;
}
/**
* Returns the largest half-precision float value toward positive infinity
* less than or equal to the specified half-precision float value.
* Special values are handled in the following ways:
* <ul>
* <li>If the specified half-precision float is NaN, the result is NaN</li>
* <li>If the specified half-precision float is infinity (negative or positive),
* the result is infinity (with the same sign)</li>
* <li>If the specified half-precision float is zero (negative or positive),
* the result is zero (with the same sign)</li>
* </ul>
*
* @param h A half-precision float value
* @return The largest half-precision float value toward positive infinity
* less than or equal to the specified half-precision float value
*/
public static @HalfFloat short floor(@HalfFloat short h) {
int bits = h & 0xffff;
int e = bits & 0x7fff;
int result = bits;
if (e < 0x3c00) {
result &= FP16_SIGN_MASK;
result |= 0x3c00 & (bits > 0x8000 ? 0xffff : 0x0);
} else if (e < 0x6400) {
e = 25 - (e >> 10);
int mask = (1 << e) - 1;
result += mask & -(bits >> 15);
result &= ~mask;
}
return (short) result;
}
/**
* Returns the truncated half-precision float value of the specified
* half-precision float value. Special values are handled in the following ways:
* <ul>
* <li>If the specified half-precision float is NaN, the result is NaN</li>
* <li>If the specified half-precision float is infinity (negative or positive),
* the result is infinity (with the same sign)</li>
* <li>If the specified half-precision float is zero (negative or positive),
* the result is zero (with the same sign)</li>
* </ul>
*
* @param h A half-precision float value
* @return The truncated half-precision float value of the specified
* half-precision float value
*/
public static @HalfFloat short trunc(@HalfFloat short h) {
int bits = h & 0xffff;
int e = bits & 0x7fff;
int result = bits;
if (e < 0x3c00) {
result &= FP16_SIGN_MASK;
} else if (e < 0x6400) {
e = 25 - (e >> 10);
int mask = (1 << e) - 1;
result &= ~mask;
}
return (short) result;
}
/**
* Returns the smaller of two half-precision float values (the value closest
* to negative infinity). Special values are handled in the following ways:
* <ul>
* <li>If either value is NaN, the result is NaN</li>
* <li>{@link #NEGATIVE_ZERO} is smaller than {@link #POSITIVE_ZERO}</li>
* </ul>
*
* @param x The first half-precision value
* @param y The second half-precision value
* @return The smaller of the two specified half-precision values
*/
public static @HalfFloat short min(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return NaN;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return NaN;
if ((x & FP16_COMBINED) == 0 && (y & FP16_COMBINED) == 0) {
return (x & FP16_SIGN_MASK) != 0 ? x : y;
}
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) <
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff) ? x : y;
}
/**
* Returns the larger of two half-precision float values (the value closest
* to positive infinity). Special values are handled in the following ways:
* <ul>
* <li>If either value is NaN, the result is NaN</li>
* <li>{@link #POSITIVE_ZERO} is greater than {@link #NEGATIVE_ZERO}</li>
* </ul>
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return The larger of the two specified half-precision values
*/
public static @HalfFloat short max(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return NaN;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return NaN;
if ((x & FP16_COMBINED) == 0 && (y & FP16_COMBINED) == 0) {
return (x & FP16_SIGN_MASK) != 0 ? y : x;
}
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) >
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff) ? x : y;
}
/**
* Returns true if the first half-precision float value is less (smaller
* toward negative infinity) than the second half-precision float value.
* If either of the values is NaN, the result is false.
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return True if x is less than y, false otherwise
*/
public static boolean less(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) <
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff);
}
/**
* Returns true if the first half-precision float value is less (smaller
* toward negative infinity) than or equal to the second half-precision
* float value. If either of the values is NaN, the result is false.
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return True if x is less than or equal to y, false otherwise
*/
public static boolean lessEquals(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) <=
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff);
}
/**
* Returns true if the first half-precision float value is greater (larger
* toward positive infinity) than the second half-precision float value.
* If either of the values is NaN, the result is false.
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return True if x is greater than y, false otherwise
*/
public static boolean greater(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) >
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff);
}
/**
* Returns true if the first half-precision float value is greater (larger
* toward positive infinity) than or equal to the second half-precision float
* value. If either of the values is NaN, the result is false.
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return True if x is greater than y, false otherwise
*/
public static boolean greaterEquals(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
return ((x & FP16_SIGN_MASK) != 0 ? 0x8000 - (x & 0xffff) : x & 0xffff) >=
((y & FP16_SIGN_MASK) != 0 ? 0x8000 - (y & 0xffff) : y & 0xffff);
}
/**
* Returns true if the two half-precision float values are equal.
* If either of the values is NaN, the result is false. {@link #POSITIVE_ZERO}
* and {@link #NEGATIVE_ZERO} are considered equal.
*
* @param x The first half-precision value
* @param y The second half-precision value
*
* @return True if x is equal to y, false otherwise
*/
public static boolean equals(@HalfFloat short x, @HalfFloat short y) {
if ((x & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
if ((y & FP16_COMBINED) > FP16_EXPONENT_MAX) return false;
return x == y || ((x | y) & FP16_COMBINED) == 0;
}
/**
* Returns the sign of the specified half-precision float.
*
* @param h A half-precision float value
* @return 1 if the value is positive, -1 if the value is negative
*/
public static int getSign(@HalfFloat short h) {
return (h & FP16_SIGN_MASK) == 0 ? 1 : -1;
}
/**
* Returns the unbiased exponent used in the representation of
* the specified half-precision float value. if the value is NaN
* or infinite, this* method returns {@link #MAX_EXPONENT} + 1.
* If the argument is 0 or a subnormal representation, this method
* returns {@link #MIN_EXPONENT} - 1.
*
* @param h A half-precision float value
* @return The unbiased exponent of the specified value
*/
public static int getExponent(@HalfFloat short h) {
return ((h >>> FP16_EXPONENT_SHIFT) & FP16_EXPONENT_MASK) - FP16_EXPONENT_BIAS;
}
/**
* Returns the significand, or mantissa, used in the representation
* of the specified half-precision float value.
*
* @param h A half-precision float value
* @return The significand, or significand, of the specified vlaue
*/
public static int getSignificand(@HalfFloat short h) {
return h & FP16_SIGNIFICAND_MASK;
}
/**
* Returns true if the specified half-precision float value represents
* infinity, false otherwise.
*
* @param h A half-precision float value
* @return True if the value is positive infinity or negative infinity,
* false otherwise
*/
public static boolean isInfinite(@HalfFloat short h) {
return (h & FP16_COMBINED) == FP16_EXPONENT_MAX;
}
/**
* Returns true if the specified half-precision float value represents
* a Not-a-Number, false otherwise.
*
* @param h A half-precision float value
* @return True if the value is a NaN, false otherwise
*/
public static boolean isNaN(@HalfFloat short h) {
return (h & FP16_COMBINED) > FP16_EXPONENT_MAX;
}
/**
* Returns true if the specified half-precision float value is normalized
* (does not have a subnormal representation). If the specified value is
* {@link #POSITIVE_INFINITY}, {@link #NEGATIVE_INFINITY},
* {@link #POSITIVE_ZERO}, {@link #NEGATIVE_ZERO}, NaN or any subnormal
* number, this method returns false.
*
* @param h A half-precision float value
* @return True if the value is normalized, false otherwise
*/
public static boolean isNormalized(@HalfFloat short h) {
return (h & FP16_EXPONENT_MAX) != 0 && (h & FP16_EXPONENT_MAX) != FP16_EXPONENT_MAX;
}
/**
* <p>Converts the specified half-precision float value into a
* single-precision float value. The following special cases are handled:</p>
* <ul>
* <li>If the input is {@link #NaN}, the returned value is {@link Float#NaN}</li>
* <li>If the input is {@link #POSITIVE_INFINITY} or
* {@link #NEGATIVE_INFINITY}, the returned value is respectively
* {@link Float#POSITIVE_INFINITY} or {@link Float#NEGATIVE_INFINITY}</li>
* <li>If the input is 0 (positive or negative), the returned value is +/-0.0f</li>
* <li>Otherwise, the returned value is a normalized single-precision float value</li>
* </ul>
*
* @param h The half-precision float value to convert to single-precision
* @return A normalized single-precision float value
*/
public static float toFloat(@HalfFloat short h) {
int bits = h & 0xffff;
int s = bits & FP16_SIGN_MASK;
int e = (bits >>> FP16_EXPONENT_SHIFT) & FP16_EXPONENT_MASK;
int m = (bits ) & FP16_SIGNIFICAND_MASK;
int outE = 0;
int outM = 0;
if (e == 0) { // Denormal or 0
if (m != 0) {
// Convert denorm fp16 into normalized fp32
float o = Float.intBitsToFloat(FP32_DENORMAL_MAGIC + m);
o -= FP32_DENORMAL_FLOAT;
return s == 0 ? o : -o;
}
} else {
outM = m << 13;
if (e == 0x1f) { // Infinite or NaN
outE = 0xff;
if (outM != 0) { // SNaNs are quieted
outM |= FP32_QNAN_MASK;
}
} else {
outE = e - FP16_EXPONENT_BIAS + FP32_EXPONENT_BIAS;
}
}
int out = (s << 16) | (outE << FP32_EXPONENT_SHIFT) | outM;
return Float.intBitsToFloat(out);
}
/**
* <p>Converts the specified single-precision float value into a
* half-precision float value. The following special cases are handled:</p>
* <ul>
* <li>If the input is NaN (see {@link Float#isNaN(float)}), the returned
* value is {@link #NaN}</li>
* <li>If the input is {@link Float#POSITIVE_INFINITY} or
* {@link Float#NEGATIVE_INFINITY}, the returned value is respectively
* {@link #POSITIVE_INFINITY} or {@link #NEGATIVE_INFINITY}</li>
* <li>If the input is 0 (positive or negative), the returned value is
* {@link #POSITIVE_ZERO} or {@link #NEGATIVE_ZERO}</li>
* <li>If the input is a less than {@link #MIN_VALUE}, the returned value
* is flushed to {@link #POSITIVE_ZERO} or {@link #NEGATIVE_ZERO}</li>
* <li>If the input is a less than {@link #MIN_NORMAL}, the returned value
* is a denorm half-precision float</li>
* <li>Otherwise, the returned value is rounded to the nearest
* representable half-precision float value</li>
* </ul>
*
* @param f The single-precision float value to convert to half-precision
* @return A half-precision float value
*/
@SuppressWarnings("StatementWithEmptyBody")
public static @HalfFloat short toHalf(float f) {
int bits = Float.floatToRawIntBits(f);
int s = (bits >>> FP32_SIGN_SHIFT );
int e = (bits >>> FP32_EXPONENT_SHIFT) & FP32_EXPONENT_MASK;
int m = (bits ) & FP32_SIGNIFICAND_MASK;
int outE = 0;
int outM = 0;
if (e == 0xff) { // Infinite or NaN
outE = 0x1f;
outM = m != 0 ? 0x200 : 0;
} else {
e = e - FP32_EXPONENT_BIAS + FP16_EXPONENT_BIAS;
if (e >= 0x1f) { // Overflow
outE = 0x31;
} else if (e <= 0) { // Underflow
if (e < -10) {
// The absolute fp32 value is less than MIN_VALUE, flush to +/-0
} else {
// The fp32 value is a normalized float less than MIN_NORMAL,
// we convert to a denorm fp16
m = (m | 0x800000) >> (1 - e);
if ((m & 0x1000) != 0) m += 0x2000;
outM = m >> 13;
}
} else {
outE = e;
outM = m >> 13;
if ((m & 0x1000) != 0) {
// Round to nearest "0.5" up
int out = (outE << FP16_EXPONENT_SHIFT) | outM;
out++;
return (short) (out | (s << FP16_SIGN_SHIFT));
}
}
}
return (short) ((s << FP16_SIGN_SHIFT) | (outE << FP16_EXPONENT_SHIFT) | outM);
}
/**
* Returns a {@code Half} instance representing the specified
* half-precision float value.
*
* @param h A half-precision float value
* @return a {@code Half} instance representing {@code h}
*/
public static @NonNull Half valueOf(@HalfFloat short h) {
return new Half(h);
}
/**
* Returns a {@code Half} instance representing the specified float value.
*
* @param f A float value
* @return a {@code Half} instance representing {@code f}
*/
public static @NonNull Half valueOf(float f) {
return new Half(f);
}
/**
* Returns a {@code Half} instance representing the specified string value.
* Calling this method is equivalent to calling
* <code>toHalf(Float.parseString(h))</code>. See {@link Float#valueOf(String)}
* for more information on the format of the string representation.
*
* @param s The string to be parsed
* @return a {@code Half} instance representing {@code h}
* @throws NumberFormatException if the string does not contain a parsable
* half-precision float value
*/
public static @NonNull Half valueOf(@NonNull String s) {
return new Half(s);
}
/**
* Returns the half-precision float value represented by the specified string.
* Calling this method is equivalent to calling
* <code>toHalf(Float.parseString(h))</code>. See {@link Float#valueOf(String)}
* for more information on the format of the string representation.
*
* @param s The string to be parsed
* @return A half-precision float value represented by the string
* @throws NumberFormatException if the string does not contain a parsable
* half-precision float value
*/
public static @HalfFloat short parseHalf(@NonNull String s) throws NumberFormatException {
return toHalf(Float.parseFloat(s));
}
/**
* Returns a string representation of the specified half-precision
* float value. Calling this method is equivalent to calling
* <code>Float.toString(toFloat(h))</code>. See {@link Float#toString(float)}
* for more information on the format of the string representation.
*
* @param h A half-precision float value
* @return A string representation of the specified value
*/
@NonNull
public static String toString(@HalfFloat short h) {
return Float.toString(toFloat(h));
}
/**
* <p>Returns a hexadecimal string representation of the specified half-precision
* float value. If the value is a NaN, the result is <code>"NaN"</code>,
* otherwise the result follows this format:</p>
* <ul>
* <li>If the sign is positive, no sign character appears in the result</li>
* <li>If the sign is negative, the first character is <code>'-'</code></li>
* <li>If the value is inifinity, the string is <code>"Infinity"</code></li>
* <li>If the value is 0, the string is <code>"0x0.0p0"</code></li>
* <li>If the value has a normalized representation, the exponent and
* significand are represented in the string in two fields. The significand
* starts with <code>"0x1."</code> followed by its lowercase hexadecimal
* representation. Trailing zeroes are removed unless all digits are 0, then
* a single zero is used. The significand representation is followed by the
* exponent, represented by <code>"p"</code>, itself followed by a decimal
* string of the unbiased exponent</li>
* <li>If the value has a subnormal representation, the significand starts
* with <code>"0x0."</code> followed by its lowercase hexadecimal
* representation. Trailing zeroes are removed unless all digits are 0, then
* a single zero is used. The significand representation is followed by the
* exponent, represented by <code>"p-14"</code></li>
* </ul>
*
* @param h A half-precision float value
* @return A hexadecimal string representation of the specified value
*/
@NonNull
public static String toHexString(@HalfFloat short h) {
StringBuilder o = new StringBuilder();
int bits = h & 0xffff;
int s = (bits >>> FP16_SIGN_SHIFT );
int e = (bits >>> FP16_EXPONENT_SHIFT) & FP16_EXPONENT_MASK;
int m = (bits ) & FP16_SIGNIFICAND_MASK;
if (e == 0x1f) { // Infinite or NaN
if (m == 0) {
if (s != 0) o.append('-');
o.append("Infinity");
} else {
o.append("NaN");
}
} else {
if (s == 1) o.append('-');
if (e == 0) {
if (m == 0) {
o.append("0x0.0p0");
} else {
o.append("0x0.");
String significand = Integer.toHexString(m);
o.append(significand.replaceFirst("0{2,}$", ""));
o.append("p-14");
}
} else {
o.append("0x1.");
String significand = Integer.toHexString(m);
o.append(significand.replaceFirst("0{2,}$", ""));
o.append('p');
o.append(Integer.toString(e - FP16_EXPONENT_BIAS));
}
}
return o.toString();
}
}
|